[v7] ethdev: fix strict aliasing lead to link cannot be up
Checks
Commit Message
Fix a problem introduced by a compiler upgrade (from gcc10 to gcc12.3),
which will lead the hns3 NIC can't link up. The root cause is strict
aliasing violation in rte_eth_linkstatus_set() with hns3 driver, see
[1] for more details.
This commit use union to avoid such aliasing violation.
Note: DPDK CI report compiler error (see [2] for more details):
../drivers/net/cxgbe/cxgbe_ethdev.c:214:9: error: missing braces around
initializer [-Werror=missing-braces]
struct rte_eth_link new_link = { 0 };
The same error with qos_sched example:
../examples/qos_sched/init.c:338:10: error: missing braces around
initializer [-Werror=missing-braces]
struct rte_eth_link link = {0};
So this commit replace { 0 } with memset in cxgbe and qos_sched.
[1] Strict aliasing problem with rte_eth_linkstatus_set()
https://marc.info/?l=dpdk-dev&m=171274148514777&w=3
[2] https://mails.dpdk.org/archives/test-report/2024-April/637966.html
Cc: stable@dpdk.org
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Dengdui Huang <huangdengdui@huawei.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
---
v7: add rte_atomic_store_explicit to get link status which address
Morten's comment, and add Morten's acked-by.
v6: fix DPDK CI compiler error with qos_sched.
v5: fix DPDK CI compiler error due to GCC extension [-Werror=pedantic]
v4: fix DPDK CI compiler error with cxgbe.
v3: fix checkpatch warning "missing --in-reply-to".
v2: add RTE_ATOMIC(uint64_t) wrap which address Morten's comment.
---
drivers/net/cxgbe/cxgbe_ethdev.c | 3 ++-
examples/qos_sched/init.c | 3 ++-
lib/ethdev/ethdev_driver.h | 25 +++++++++----------------
lib/ethdev/rte_ethdev.h | 17 +++++++++++------
4 files changed, 24 insertions(+), 24 deletions(-)
Comments
On 4/18/2024 8:28 AM, Chengwen Feng wrote:
> @@ -1701,12 +1696,10 @@ static inline void
> rte_eth_linkstatus_get(const struct rte_eth_dev *dev,
> struct rte_eth_link *link)
> {
> - RTE_ATOMIC(uint64_t) *src = (uint64_t __rte_atomic *)&(dev->data->dev_link);
> - uint64_t *dst = (uint64_t *)link;
> -
> - RTE_BUILD_BUG_ON(sizeof(*link) != sizeof(uint64_t));
> -
> - *dst = rte_atomic_load_explicit(src, rte_memory_order_seq_cst);
> + rte_atomic_store_explicit(&link->val64,
> + rte_atomic_load_explicit(&dev->data->dev_link.val64,
> + rte_memory_order_seq_cst),
> + rte_memory_order_seq_cst);
>
Hi Chengwen,
Overall looks good to me, but to increase readability what do you think
to extract function call in the function param, like
```
struct rte_eth_link new_link;
new_link->val64 = rte_atomic_load_explicit(..dev_link.val64, ..);
rte_atomic_store_explicit(&link->val64, new_link->val64, ..);
```
On 4/18/2024 8:28 AM, Chengwen Feng wrote:
> Fix a problem introduced by a compiler upgrade (from gcc10 to gcc12.3),
> which will lead the hns3 NIC can't link up. The root cause is strict
> aliasing violation in rte_eth_linkstatus_set() with hns3 driver, see
> [1] for more details.
>
> This commit use union to avoid such aliasing violation.
>
> Note: DPDK CI report compiler error (see [2] for more details):
> ../drivers/net/cxgbe/cxgbe_ethdev.c:214:9: error: missing braces around
> initializer [-Werror=missing-braces]
> struct rte_eth_link new_link = { 0 };
> The same error with qos_sched example:
> ../examples/qos_sched/init.c:338:10: error: missing braces around
> initializer [-Werror=missing-braces]
> struct rte_eth_link link = {0};
> So this commit replace { 0 } with memset in cxgbe and qos_sched.
>
As this commit is already fixing the build errors, not sure if there is
a value to provide reference to errors, you can briefly describe change
something like:
"The impacted components have been adapted to the struct change."
> [1] Strict aliasing problem with rte_eth_linkstatus_set()
> https://marc.info/?l=dpdk-dev&m=171274148514777&w=3
>
I wasn't aware marc.info, but for consistency you can use DPDK mail list
archive, inbox.dpdk.org, like:
https://inbox.dpdk.org/dev/8175c905-e661-b910-7f20-59b6ab605c38@huawei.com/
> [2] https://mails.dpdk.org/archives/test-report/2024-April/637966.html
>
> Cc: stable@dpdk.org
>
> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
> Signed-off-by: Dengdui Huang <huangdengdui@huawei.com>
> Acked-by: Morten Brørup <mb@smartsharesystems.com>
Hi Ferruh,
On 2024/4/19 23:25, Ferruh Yigit wrote:
> On 4/18/2024 8:28 AM, Chengwen Feng wrote:
>> Fix a problem introduced by a compiler upgrade (from gcc10 to gcc12.3),
>> which will lead the hns3 NIC can't link up. The root cause is strict
>> aliasing violation in rte_eth_linkstatus_set() with hns3 driver, see
>> [1] for more details.
>>
>> This commit use union to avoid such aliasing violation.
>>
>> Note: DPDK CI report compiler error (see [2] for more details):
>> ../drivers/net/cxgbe/cxgbe_ethdev.c:214:9: error: missing braces around
>> initializer [-Werror=missing-braces]
>> struct rte_eth_link new_link = { 0 };
>> The same error with qos_sched example:
>> ../examples/qos_sched/init.c:338:10: error: missing braces around
>> initializer [-Werror=missing-braces]
>> struct rte_eth_link link = {0};
>> So this commit replace { 0 } with memset in cxgbe and qos_sched.
>>
>
> As this commit is already fixing the build errors, not sure if there is
> a value to provide reference to errors, you can briefly describe change
> something like:
> "The impacted components have been adapted to the struct change."
ok
>
>
>> [1] Strict aliasing problem with rte_eth_linkstatus_set()
>> https://marc.info/?l=dpdk-dev&m=171274148514777&w=3
>>
>
> I wasn't aware marc.info, but for consistency you can use DPDK mail list
> archive, inbox.dpdk.org, like:
> https://inbox.dpdk.org/dev/8175c905-e661-b910-7f20-59b6ab605c38@huawei.com/
ok
both done in v8.
Thanks
>
>> [2] https://mails.dpdk.org/archives/test-report/2024-April/637966.html
>>
>> Cc: stable@dpdk.org
>>
>> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
>> Signed-off-by: Dengdui Huang <huangdengdui@huawei.com>
>> Acked-by: Morten Brørup <mb@smartsharesystems.com>
>
> .
>
@@ -211,9 +211,9 @@ int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
unsigned int i, work_done, budget = 32;
struct link_config *lc = &pi->link_cfg;
struct adapter *adapter = pi->adapter;
- struct rte_eth_link new_link = { 0 };
u8 old_link = pi->link_cfg.link_ok;
struct sge *s = &adapter->sge;
+ struct rte_eth_link new_link;
for (i = 0; i < CXGBE_LINK_STATUS_POLL_CNT; i++) {
if (!s->fw_evtq.desc)
@@ -232,6 +232,7 @@ int cxgbe_dev_link_update(struct rte_eth_dev *eth_dev,
rte_delay_ms(CXGBE_LINK_STATUS_POLL_MS);
}
+ memset(&new_link, 0, sizeof(new_link));
new_link.link_status = cxgbe_force_linkup(adapter) ?
RTE_ETH_LINK_UP : pi->link_cfg.link_ok;
new_link.link_autoneg = (lc->link_caps & FW_PORT_CAP32_ANEG) ? 1 : 0;
@@ -335,7 +335,7 @@ int app_init(void)
for(i = 0; i < nb_pfc; i++) {
uint32_t socket = rte_lcore_to_socket_id(qos_conf[i].rx_core);
struct rte_ring *ring;
- struct rte_eth_link link = {0};
+ struct rte_eth_link link;
int retry_count = 100, retry_delay = 100; /* try every 100ms for 10 sec */
snprintf(ring_name, MAX_NAME_LEN, "ring-%u-%u", i, qos_conf[i].rx_core);
@@ -367,6 +367,7 @@ int app_init(void)
app_init_port(qos_conf[i].rx_port, qos_conf[i].mbuf_pool);
app_init_port(qos_conf[i].tx_port, qos_conf[i].mbuf_pool);
+ memset(&link, 0, sizeof(link));
rte_eth_link_get(qos_conf[i].tx_port, &link);
if (link.link_status == 0)
printf("Waiting for link on port %u\n", qos_conf[i].tx_port);
@@ -1674,18 +1674,13 @@ static inline int
rte_eth_linkstatus_set(struct rte_eth_dev *dev,
const struct rte_eth_link *new_link)
{
- RTE_ATOMIC(uint64_t) *dev_link = (uint64_t __rte_atomic *)&(dev->data->dev_link);
- union {
- uint64_t val64;
- struct rte_eth_link link;
- } orig;
-
- RTE_BUILD_BUG_ON(sizeof(*new_link) != sizeof(uint64_t));
+ struct rte_eth_link old_link;
- orig.val64 = rte_atomic_exchange_explicit(dev_link, *(const uint64_t *)new_link,
- rte_memory_order_seq_cst);
+ old_link.val64 = rte_atomic_exchange_explicit(&dev->data->dev_link.val64,
+ new_link->val64,
+ rte_memory_order_seq_cst);
- return (orig.link.link_status == new_link->link_status) ? -1 : 0;
+ return (old_link.link_status == new_link->link_status) ? -1 : 0;
}
/**
@@ -1701,12 +1696,10 @@ static inline void
rte_eth_linkstatus_get(const struct rte_eth_dev *dev,
struct rte_eth_link *link)
{
- RTE_ATOMIC(uint64_t) *src = (uint64_t __rte_atomic *)&(dev->data->dev_link);
- uint64_t *dst = (uint64_t *)link;
-
- RTE_BUILD_BUG_ON(sizeof(*link) != sizeof(uint64_t));
-
- *dst = rte_atomic_load_explicit(src, rte_memory_order_seq_cst);
+ rte_atomic_store_explicit(&link->val64,
+ rte_atomic_load_explicit(&dev->data->dev_link.val64,
+ rte_memory_order_seq_cst),
+ rte_memory_order_seq_cst);
}
/**
@@ -332,12 +332,17 @@ struct rte_eth_stats {
/**
* A structure used to retrieve link-level information of an Ethernet port.
*/
-__extension__
-struct __rte_aligned(8) rte_eth_link { /**< aligned for atomic64 read/write */
- uint32_t link_speed; /**< RTE_ETH_SPEED_NUM_ */
- uint16_t link_duplex : 1; /**< RTE_ETH_LINK_[HALF/FULL]_DUPLEX */
- uint16_t link_autoneg : 1; /**< RTE_ETH_LINK_[AUTONEG/FIXED] */
- uint16_t link_status : 1; /**< RTE_ETH_LINK_[DOWN/UP] */
+struct rte_eth_link {
+ union {
+ RTE_ATOMIC(uint64_t) val64; /**< used for atomic64 read/write */
+ __extension__
+ struct {
+ uint32_t link_speed; /**< RTE_ETH_SPEED_NUM_ */
+ uint16_t link_duplex : 1; /**< RTE_ETH_LINK_[HALF/FULL]_DUPLEX */
+ uint16_t link_autoneg : 1; /**< RTE_ETH_LINK_[AUTONEG/FIXED] */
+ uint16_t link_status : 1; /**< RTE_ETH_LINK_[DOWN/UP] */
+ };
+ };
};
/**@{@name Link negotiation